Model-based Optimisation of Diesel Engine Systems

Because of their high fuel efficiency, modern Diesel engines are nowadays used in a wide range of applications. However, they are also responsible for a large amount of harmful emissions, namely particulate matter (PM) and nitrous oxides (NOx). As a consequence, emission legislations are being tightened gradually. State-of-the-art technologies, such as sequential dual-stage turbo charging, common rail injection, exhaust-gas recirculation (EGR) and exhaust-gas after-treatment make it possible to comply with the tightening regulations while maintaining or even improving the fuel efficiency.

These technologies add new degrees of freedom to the engine, demanding novel strategies for the development of the engine control. Up to now, the control structures were rather simple, with a strong dependency on measured maps. Due to the many degrees of freedom, this approach is not sensible because of the limited performance and the excessive need for measurements. Furthermore, depending on the application and the market, a manufacturer may use several different engine configurations. Therefore, model-based control strategies need to be used to handle the sheer complexity of a modern Diesel engine.

The project focuses on the following subtasks:

• Air-path control: Control of the cylinder charge using the turbo chargers and the EGR
• Injection control: Control of the pollutant emissions using the fuel injection
• Supervisory control: Realization of an optimal operating strategy using model-based optimization

In a first stage, this project has focused on the control of the cylinder charge. The turbo chargers deliver fresh air which mixes in the intake manifold with the recirculated exhaust gas. This gas mixture is aspirated by the cylinder and used for the combustion of the injected fuel. Both the amount and the composition of the aspirated gas mixture are crucial for the combustion. Together with the fuel injection they are the main influences on the fuel economy and the pollutant emissions, making fast and precise control of these quantities essential. However, controlling the turbo chargers and the EGR in order to supply the desired gas mixture is a complicated task because the system has several disadvantageous properties. For example, the turbo chargers and the EGR have strong influence on each other and their behavior depends strongly on the operating point of the engine. In order to handle these properties, they have to be characterized first. A system analysis is performed for the models of various engines of different size and air-path configuration. This analysis reveals that the problematic system properties have a common source for all considered engine configurations. This knowledge allows us to design a control structure, which can handle these properties for various engine configurations.

In a second stage, the project focused on the control of the injection. The cylinder charge constitutes the boundary condition for the combustion. However, the air-path controller is not perfect and especially during transients, the cylinder charge can differ from the optimal value. Whenever such an offset occurs, the fuel injection needs to be adapted in order to prevent temporary peaks in the pollutant emissions. A model-based estimation of the cylinder charge is used to calculate the necessary adaptation of the injection parameters, such that emission peaks are prevented.

In the last stage, the project focuses on supervisory control of the engine. For a given driving cycle, legislation prescribes an upper limit for the total amount of pollutant emissions. Generally, there is a trade-off between the fuel consumption and the pollutant emissions. This trade-off renders a simultaneous reduction of the fuel consumption and the pollutant emissions difficult. An operating strategy is sought, which minimizes the fuel consumption while keeping the pollutant emissions on the allowed level. The challenge is to know when it is best to save pollutant emissions at the cost of increased fuel consumption, and when to accept increased pollutant emissions in order to achieve lower fuel consumption. The supervisory controller uses an engine model to carry out an online optimization to solve this problem.

In response to the increasingly stringent emission regulations and a demand for ever lower fuel consumption, diesel engines have become complex systems. The exploitation of any leftover potential during transient operation is crucial. However, even an experienced calibration engineer cannot conceive all the dynamic cross couplings between the many actuators. Therefore, a highly iterative procedure is required to obtain a single engine calibration, which in turn causes a high demand for test-bench time. To alleviate this dilemma, the calibration process has to be automated by relying on model-based dynamic optimization. Four points are critical to implement such an approach:

• Optimisation-oriented models for the air path and for the in-cylinder processes of diesel engines need to be derived. They need to be quantitatively accurate, fast, and smooth. Moreover, the full actuator ranges and the entire engine operating range has to be covered, as well as physically plausible extrapolation needs to be provided.
• A sound problem formulation has to be derived, which always requires to trade-off the engineering meaningfulness against the numerical properties.
• Efficient numerical methods for the solution of the resulting large-scale optimal control problem have to be developed. They can be based either on a subdivision of the long time horizon or on large-scale algorithms. Both approaches require an efficient, customary implementation of the derivative generation that exploits the problem sparsity and applies parallel computing. Moreover, concepts such as an iterative mesh refinement and warm starting of the solver increase the performance of the optimization process.
• Ways of utilizing the optimal solutions have to be found. The results are non-causal and only valid for the driving profile considered during the optimization. However, if the driving profile covers the entire relevant operating range of the engine, implications for a causal control strategy can be derived. Moreover, maps for the combustion actuators as well as optimal feedforward and reference quantities for feedback control may be fitted to the optimal data.

A fully automated model-based calibration of the engine control unit was performed for an engine without EGR. For an engine with EGR, it was shown how optimal control can be utilised to develop a control strategy that provides an optimal transient fuel-NOx tradeoff. An experimental validation has been executed, and indicates that these considerations accurately transfer to the real engines.

In spite of possessing high efficiencies, Diesel engines produce high NOx emissions. To meet the ever stringent emission regulations, urea selective catalytic reduction (SCR) system is fitted on the latest heavy-duty Diesel engines. In this system, urea water solution is injected to generate ammonia, which then reduces NOx to harmless nitrogen and water. The use of SCR is preferred over many other systems primarily for its high deNOx efficiency. With this high efficiency, it further offers the possibility to remove the exhaust gas recirculation (EGR) system. The removal can result in a reduction in both fuel consumption and PM emissions.

To maximize the deNOx efficiency of SCR systems, a proper urea dosing controller has to be designed. The main control challenges are:

• Ammonia slip: Intuitively, to increase NOx conversion rate, more urea needs to be injected. However, this increases the risk of ammonia slip. In the latest emission regulations, the concentration of ammonia emission is also regulated. A well-design controller is thus needed to maximize the deNOx efficiency while keeping the ammonia slip under the regulatory level for both static and transient operations.
• NOx sensor cross-sensitivity: To feedback control the NOx emissions, a measurement on NOx concentration at SCR outlet is necessary. However, the commercial NOx sensor is cross-sensitive to NH3. Therefore, a method has to be developed to deal with this cross-sensitivity issue.
• Calibration effort: The SCR systems are often used in combination with different engines for various emission regulations. The calibration effort therefore is great, if each combination has to be calibrated at the test bench individually. In this case, a model-based approach is favored to minimize the calibration effort.

The objective of our project is to design an optimal urea dosing controller to fulfill the NOx regulations with minimal calibration effort.

To reach this objective, the following fields of study are of great interests:

• Modeling and control of SCR systems
• Engine exhaust temperature control
• Integrated emission control of Diesel engine and SCR system